The contributions of this thesis are three-fold: it proposes a new deep learning-based approach for inferring GUI rendering states, develops a high-quality dataset to support fine-tuning models, and ...